One of the most salient aspects of morphological systems is the degree to which they appear to vary across languages. This variation has been interpreted as presenting a fundamental challenge to any unified morphological model (Matthews 1972:156) or even as evidence that morphology is somehow “unnatural” (Aronoff 1998:413). However, a framework of analysis that focuses more on implicational relations and discriminative contrasts between units than on the units themselves suggests a basis for a general morphological model that can accommodate the range of attested variation across languages.
Typological and theoretical overfit
The shift from a unit-based to a relation-based perspective helps to reduce the risk of fitting a morphological model too closely to specific patterns of form variation. This problem arises in a particularly acute form in the essentially procrustean procedures of morphemic analysis discussed in Chapter 2. The realizational WP tradition initiated by Matthews (1972) has compiled a catalogue of patterns that defy description in morphemic terms. As critiques of morphemic analysis, these studies are highly effective and leave no doubt that in at least some languages, the hypothesized “matching between minimal ‘sames’ of ‘form’ (morphs) and ‘meaning’ (morphemes)... does not exist” (Matthews 1972:124). However, this tradition has been somewhat less successful in formulating a cohesive alternative conception of morphological organization above the implementation level. Demonstrating that a morphemic correspondence cannot be established in all languages does not entail that the correspondence must be incidental in those where it does appear to hold, or that non-morphemic morphotactic units play no grammatical role. Realizational approaches have also tended to treat the identification of ‘non-morphemic’ patterns as an end in itself and have been more inclined to classify these patterns as ‘residue’ or ‘noise’ than to try to determine their function.
There are multiple reasons for the lack of a unifying set of organizational principles within realizational approaches. In the empirical domain, the disunity derives in part from the fact that these accounts often focus on patterns that are recalcitrant for morphemic models and recalcitrant patterns tend not to comprise a coherent class. From an institutional standpoint, it is relevant that proponents of realizational models form less of an integrated school or community than the Bloomfieldians or their successors. Hence there is a greater diversity of perspectives represented in the realizational WP tradition. These views range from scepticism regarding any claims to universality (Matthews 1972), to theory-specific claims about formal universals (Stump 2001) or default inheritance mechanisms (Brown and Hippisley 2012), to the endorsement of an essentially diachronic perspective (Anderson 1992) and even to assertions that morphology is a “pathology of language” (Aronoff 1994:413).
The possibility that typological variation may determine a variety of models, each optimized for specific patterns, is taken most seriously by Matthews:
Finally, it has become clear at least that different languages raise quite different problems in morphological analysis. It is therefore possible that they also require quite different sorts of description. (Matthews 1972:156)
From this standpoint, there is no principled reason to expect a ‘one size fits all’ model of morphology. Instead, the best description of cross-linguistic variation may be provided by a theory that makes available a range of different models, each tuned to particular types of patterns or languages. There maybe ‘formal universals’ (in the sense of Chomsky 1965) that characterize the class of models defined by a morphological theory. But the search for more substantive universal principles or constraints, of the kind represented by morphemic biuniqueness, may turn out to be futile or even misconceived.
The Pamnian Determinism Hypothesis proposed by Stump (2001: 23) provides a clear example of the type of theory-internal constraint formulated within the realizational WP tradition. This hypothesis specifies that “[competition among members of the same rule block is in all cases resolved by Panini’s principle”, which requires that rules within a block must either be incompatible or ranked by relative specificity. Although this hypothesis constrains the strategies for determining rule priority within a block, the restrictions it imposes also depend on the degree of freedom in the determination of rule blocks, rule compatibility and relative specificity. If none of these components are independently fixed, it is possible to regulate nearly any alternation by appealing to a ‘specificity’-based principle. Because the hypothesis presupposes all of these components, its relevance is also limited to a narrowly circumscribed set of approaches. In particular, the hypothesis does not generalize in any straightforward way to classical WP models or even to realization- based models without rule blocks. From a learning-based perspective, it is also unexpected that a model would be highly sensitive to the relative specificity of rules and utterly insensitive to distributional information.
On a traditional understanding of the term, an ‘explanation’ for recurrent patterns and variation in and across morphological systems must lie outside those systems. A standard locus of explanatory factors is the diachronic domain, and a diachronic perspective on morphological variation is taken up by Anderson (1992). Echoing the approach to sound systems proposed by Baudouin de Courtenay (1895), Anderson suggests that morphological patterns are determined by the possible pathways of evolutionary development:
Especially in the domain of morphology, where much that we find is the product of historical change operating on originally non-morphological material, it is important to recognize that what we find is the product not only of what is possible, but also of what can come into being. (Anderson 1992:372)
This view not only offers an explanation of the space of morphological patterns and systems in terms of possible developmental trajectories, but also accounts for distributional regularities within that space. The prevalence of a particular morphological pattern will correlate not with the ‘markedness’ of the pattern in some abstract system-internal sense, but with the number of different pathways of change that can produce the pattern and the frequency of these pathways. Hence the recurrence of morphemic patterns need not reflect anything about the organization of morphological systems. Instead, this generalization (if, indeed, it is one) can be attributed to the way that new morphology tends to arise. If morphologization involves the prosodic reanalysis of formerly free elements as bound formatives, then the meaning and features of the original element may remain associated with the newly bound formative.
There can be no reasonable doubt that any genuine explanations for morphological patterns or systems must lie outside those systems and that diachronic accounts provide the most secure basis for a predictive, explanatory model. There are, however, two fundamental limitations of this perspective. The first is that the space of possible ‘next moves’ in a morphological system is much denser and less predictable than in sound systems. This difference is strikingly reflected in the fact that morphological change is not sufficiently constrained to serve as the basis for language reconstruction using the comparative method. Although cognate morphology, particularly irregular patterns, plays an important role in establishing genetic affiliation, there is no morphological counterpart of regular sound change. It maybe that emerging simulation-based methodologies (Sproat 2008) will provide a means of plotting the developmental trajectories of morphological systems, though at the moment diachronic approaches still offer more in the way of promise than results. The second limitation of a historical perspective is that it does not classify the function of the elements in an existing synchronic system, except insofar as it associates them with a potential for change along one or another dimension. It is of course possible that there is nothing more to say about a synchronic system once its origins and developmental history are sufficiently well understood. But standard accounts also assume that the patterns established within a system influence the directions of change, as reflected particularly by processes like morphological levelling and extension, which are taken to enhance system- level congruence. This is another domain in which a discriminative learning approach may offer a useful perspective on how the patterns in one part of a language guide learners’ expectations in other parts.
Before considering synchronic factors that may contribute to congruence, it is worth considering the most sceptical view about the status of organizational principles within morphological systems. This is represented by what Aronoff (1998:413) memorably terms the “disease view of morphology”:
Morphology is inherently unnatural. It’s a disease, a pathology of language. This fact is demonstrated very simply by the fact that there are languages, though not that many, that manage without it—you don’t need morphology—and by the perhaps more widely recognized fact that some languages like West Greenlandic or Navajo have morphology much worse than others do. I think it’s clear that the notion of morphologization or grammaticalization is rooted in this disease view of morphology as being inherently unnatural... (Aronoff 1998:413)
On this extreme view, morphology is essentially noise within a communication channel, a pernicious source of cross-linguistic variability and language-internal arbitrariness. The sheer ferocity of Aronoff’s assault calls for some explanation. The charge of pathology cannot simply be grounded in the fact that morphology is not universally present (even assuming that it isn’t). Plenty of linguistic features are non-universal, from consonant voicing contrasts to articles, to syntactically distinctive word order variation. In the languages in which these features are present, they tend to carry some functional load; in languages that lack these features, the load is distributed across the available resources. There is no sense in which non-universality implies pathology, either at the level of linguistic features or at the level of whole components.
The real source of pathology must be sought elsewhere. The most plausible source is the non-morphemic (rather than non-universal) nature of much morphological variation. From the observation that morphology cannot be treated as a direct conduit of grammatical meaning, Aronoff seems to conclude that morphology is therefore an unnatural feature of language. This bleak assessment reveals an insidious effect of the Post-Bloomfieldian tradition, for clearly a morphological system in which form variation did transparently mirror grammatical meaning would not be unnatural or pathological. The suggestion that naturalness is correlated with a morphemic feature-form correspondence is reinforced by the negative characterization of morphomes as ‘purely morphological, i.e. non-morphemic, elements, as mentioned in Chapter 5.2. Hence the ‘cure’ for the morphological disease lies in diagnosing the positive function that morphomic patterns play in a morphological system and understanding how this function relates to morphemic patterns.
-  As proposed in greater detail in the domain of sound patterns in the context of the model ofEvolutionary Phonology by Blevins (2004, 2006a, 20Г4).
-  Morphological patterns also figure prominently in the mass comparison descriptions promotedby Greenberg (Greenberg 1966, 1987) and in other approaches to language reconstruction and subgrouping based on surface similarities.